Ioannis Mitliagkas

Also known as Γιάννης Μητλιάγκας, Yannis Mitliagkas

Ioannis Mitliagkas is a machine learning researcher who studies optimization, dynamics and learning, asking why training works and when models generalize. After a PhD at the University of Texas at Austin in 2015 and a Stanford postdoc with Lester Mackey and Chris Ré, he showed in 2016, with Ce Zhang, Stefan Hadjis and Christopher Ré, that running SGD asynchronously can be viewed as adding a momentum-like term, and in 2017 he and Jian Zhang released the YellowFin momentum tuner. In September 2017, he joined the Université de Montréal and Mila as an assistant professor, and in 2018 he joined the first cohort of Canada CIFAR AI Chairs. His work on negative momentum for game dynamics appeared at AISTATS 2019, and in 2024 he coauthored the findings of the first NeurIPS machine unlearning competition. He is an associate professor at Montréal and a part-time staff research scientist at Google DeepMind Montréal.

Now at: Université de Montréal, Mila, Google DeepMind and Athena Research and Innovation Center In Information Communication & Knowledge Technologies
Badges: Wayfarer

Timeline

2015–2017

PhD and a Stanford postdoc

He defends his PhD thesis in electrical and computer engineering at the University of Texas at Austin in 2015. In 2015 he also starts a postdoc at Stanford University, working with Lester Mackey and Chris Ré as a postdoctoral scholar in the Departments of Statistics and Computer Science.
2016–2018

Asynchrony, momentum and YellowFin

In 2016, he coauthors a paper with Ce Zhang, Stefan Hadjis and Christopher Ré showing that running stochastic gradient descent asynchronously can be viewed as adding a momentum-like term. In 2017, he and Jian Zhang release the preprint of YellowFin, an automatic tuner for momentum and learning rate in SGD that converges in fewer iterations than Adam on ResNets and LSTMs in their experiments. YellowFin is selected for an oral presentation at SysML in 2018.
September 2017

Assistant professor in Montréal

In September 2017, he starts work as an assistant professor at the Université de Montréal and Mila, in the Department of Computer Science and Operations Research.
2018

A Canada CIFAR AI Chair

In December 2018, he joins the first cohort of Canada CIFAR AI Chairs at Mila. CIFAR renews his appointment in 2023.
2018–2019

Negative momentum for games

In 2018, he coauthors Negative Momentum for Improved Game Dynamics, which proves that alternating gradient updates are more stable than simultaneous ones and shows that a negative momentum term achieves convergence on saturating GANs; the paper appears at AISTATS 2019. He co-organizes the Smooth Games Optimization and Machine Learning workshop at NeurIPS in 2018 and 2019.
2022

Associate professor with tenure

In June 2022, he is promoted to associate professor with tenure at the Université de Montréal. In spring 2022, he is a Visiting Scientist in the Learning and Games program at the Simons Institute in Berkeley.
2024

The NeurIPS unlearning competition

In 2024, he is one of 15 authors of the findings from the first NeurIPS machine unlearning competition, which draws nearly 1,200 teams and finds that the top entries surpass existing algorithms. He spends the winter semester of 2024 on sabbatical at Archimedes in Athens.
As of October 2026

Associate professor and DeepMind scientist

He is an associate professor of computer science at the Université de Montréal, a Core Academic member of Mila and a Canada CIFAR AI Chair. He holds a part-time position as a staff research scientist at Google DeepMind Montréal, after starting part-time work at Google Brain Montréal in September 2022. He is also an affiliate researcher at Archimedes, part of the Athena Research Center in Athens.